Michael Stewart (@protokultur) header
Michael Stewart (@protokultur) profile picture

Managing Partner at @M12vc leading frontier investment in AI, gaming, semiconductors, materials science

San Francisco, CA
Joined October 2015
Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)Photo by Michael Stewart (@protokultur)
Michael Stewart retweeted
Nuns love Hermes
63
45
13
1,067
95,854
Michael Stewart retweeted
Law discovery is a scientific puzzle. The agent gets 40 read-outs from made-up instruments. Each log was generated by one hidden formula plus some noise, and the job is to discover the formula for every one. Here we benchmark on: • Straight lines and power laws (6 datasets). • Standard shapes like saturating or exponential curves (9). • Two inputs combined inside one term (8). • Kinks, caps and thresholds (7). • Oscillation and damped ringing (4). • Ones that need a dimensional or symmetry argument (6). Read more about how we quantize LLMs to generalize retention for unseen, long-horizon tasks like Law Discovery: withsyzygy.com Mach 2 Additive Medium (1.7 bit Qwen 3.8 Flash-Next): huggingface.co/SyzygyResearc…
1
2
9
459
DOCUMENT SUR 1914-18 Un an après la fin de la Première Guerre mondiale, le pilote français Jacques Trolley de Prévaux a piloté un dirigeable au dessus des champs de bataille dévastés en France. Voici ce qu'il a vu, ici en version remasterisée en couleur.🧐 (🎥@BowesChay)
64
760
27
4,048
397,089
TrendForce: AI Networking and Interconnect Market Projected to Reach $279.8 Billion in 2027, with Scale-Up Share Rising to 24% > Switches are expected to remain the largest revenue contributor, generating $105 billion in 2027 and accounting for nearly 38% of the market. > Continued migration toward higher-speed specifications is expected to lift pluggable optical transceiver revenue to $84.1 billion in 2027. > Switches, NICs/DPUs, and pluggable optical transceivers will account for approximately 88% of total AI networking and interconnect revenue. > TrendForce projects that scale-up networking's share of AI networking and interconnect revenue will grow from 11% in 2024 to 24% in 2027. Copper cables are expected to remain the dominant interconnect medium within GPU racks in 2027. > TrendForce forecasts that pluggable optical transceiver revenue will surge 76% YoY in 2027, outpacing shipment growth, with 1.6T products contributing approximately 56% of pluggable transceiver revenue. $COHR $LITE $GOOGL $NVDA $AAOI $MRVL $AVGO
8
18
1
70
8,797
Michael Stewart retweeted
At Cisco, Mike Volpi's team did 70+ acquisitions which became half of the company's revenue during the internet build-out. Now he's watching AI drive another infrastructure boom, and he thinks this cycle is even more compelling. His view: the infrastructure being built will remain useful even if some of its owners change. Hanabi Capital GP Mike Volpi joins a16z's Raghu Raghuram and Guido Appenzeller to discuss lessons from Cisco's acquisition machine, Nvidia's dominance, the opening for custom AI chips, how AI is changing data center networking, why older GPUs are holding their value, and more. 00:00 Intro 00:53 The internet buildout vs. the AI buildout 09:29 Financing the buildout, and why GPUs are holding their value 14:31 Cisco's M&A playbook, and advice for Nvidia 22:27 Making acquisitions actually integrate 25:32 When founders should sell, and how deals really happen 31:58 Mapping today's AI infrastructure 38:02 Will we see more competitive AI silicon? 43:45 Why Cisco's boom ended 46:08 Rethinking data center networking for AI 53:46 Why this is the best time to be investing YouTube: youtu.be/LI3lxzHfRrU @mavolpi @RaghuRaghuram @appenz
21
15
11
107
226,850
It has become painfully clear to me how broken even most technical peoples brains are about what's changeable in the basic equations of building a chip, computer, datacenter, computing grid. Nearly every input including the cost-perf of the cells and specific output of the panels has been scaled with no thought of space AI factories. This can all change with moderate sized investment compared to what's being deployed to scale the existing paradigms
We have signed a major supply contract with SINGULUS to create a durable option for advanced solar cell manufacturing to support more than 100 GW of equipment as we accelerate the production of our Starmind satellites in Bastrop, Texas → singulus.com/2026/11/10/spac…
2
7
4,391
Wake up babe. We are making HIT cells now.
We have signed a major supply contract with SINGULUS to create a durable option for advanced solar cell manufacturing to support more than 100 GW of equipment as we accelerate the production of our Starmind satellites in Bastrop, Texas → singulus.com/2026/11/10/spac…
80
🤯
SpaceX has ordered 100 GW/year of HJT solar manufacturing equipment from Singulus in Germany. This is for Starmind to create a sentient sun.
1
187
We know this kind of duty continues through the 41st millennium thanks to the Space Marine games
Transporte de munición de los proyectiles a bordo de un acorazado clase Iowa
2
287
Michael Stewart retweeted
After optimizing a patterning process in R&D, chipmakers use holistic lithography to keep it stable as production scales. Continuous monitoring and adjustments maintain accuracy so they can reach cost-effective volume manufacturing faster.
6
28
7
285
13,427
🤯 what the actual f?
The First 'Wireless GPU' Uses Wi-Fi to Headbutt All Those AI Subscriptions: Based on our short demo, the WiCi one still needs work before its eventual planned launch in December. It's also giving us an ounce more hope we can have a world where local compute isn't subordinated to the whims of the tech oligarchs and their pet AI. gizmodo.com/wici-one-wireles…
1
145
Looks better than the Dune movies. Storytelling may survive after all. Well done @SiebFothers
1
2
132
Michael Stewart retweeted
山梨の笛吹市で行われた #川中島合戦戦国絵巻 というイベントなのですが、合戦もあり鉄砲あり爆発あり上杉謙信の単騎突入もあり、更には甲冑武者による本気の大玉送りなどもあり見応え抜群でした。 武田信玄役の的場浩司さんと上杉謙信役の照英さんもとても格好良かったですね。
17
1,549
81
6,701
247,034
Michael Stewart retweeted
Across our projects @GoogleDeepMind, we are witnessing a genuine phase transition in the pace of scientific discovery. You will see a glimpse of this in our releases in the next few weeks.
301
784
285
9,367
1,182,290
Michael Stewart retweeted
Brute force computation is not enough for intelligence. Clever new methods are needed. The science of AI is far from finished. IMHO
429
759
523
6,549
2,582,409
"Quantum computing still has plenty of mysteries, but now there is one fewer." Below: AI Disclosures in lieu of Experimental Methods sections in manuscripts. Whoosh...
Today Physical Superintelligence PBC is announcing a proof of the Aaronson-Ambainis conjecture, one of the central open problems in quantum complexity, with major implications for quantum computing speedup. psi.inc/research/aaronson-am… Thread 👇
4
275
Invest in the deeptech systems and tech-bio nexus with the stellar performing and growing team at @TriatomicCap, see below from Managing Director @jhuber: #venturecapital #VC #careers #jobs
Thanks to amazing AI, deeptech, & techbio founders / portfolio companies, our Fund I crushed, and our Fund II landed oversubscribed. We're growing and hiring two Principals at @TriatomicCap – focused on Next-gen Compute and on Engineered Biology. We’re looking for obsessively curious people who:  . are deeply technical/scientific . think from first-principles . have conviction before others have understanding . are fully AI-pilled builders and default to creating infrastructure & products instead of spreadsheets . have real relationships with the founders, engineers, and scientists building hard stuff. Eng/Science/CS/Math/Physics with BS/MS/MD/PhD preferred. No prior VC experience required (but we won't hold it against you if founders you've backed love you). Offices in Palo Alto & SF Mission Bay. If this sounds like your dream (where you get paid to do the stuff you love anyway) – please reach out! If your inflated-equity golden handcuffs make it impossible for you right now – please point us to the smartest, most-thoughtful, most-interesting people you know, so you can live vicariously through them after they join us. :) Apply: jobs.ashbyhq.com/triatomic
2
5
409
Michael Stewart retweeted
Anthropic’s Head of Platform Engineering Katelyn Lesse explains why CPUs are now as big a capacity problem as GPUs: “And, I mean, it’s really interesting because most people are used to talking about scale or just, like, I don’t know, compute at the GPU accelerator layer and much less so at the CPU, old school layer, and it’s just as much of a problem today as the accelerator layer, which is really interesting.” “We’re using CPU across, obviously, just running our product services, running the API, running all of our apps. We’re using CPU on sandboxing, all the sandboxing that we’re doing, and then just all the things, like CI. You’d be surprised how much more code is getting written. Therefore, there’s so many more CI runs. Therefore, you need more CPU to handle that. And so on the whole, and this is not just us, this is everybody that I talk to.” “CPU for Anthropic as a whole is like a zero-sum game. We negotiate our deals with the CSPs and we get our CPU, and we have to allocate within Anthropic how much of that goes to all the things I just mentioned, including research sandboxing versus product sandboxing and otherwise. And so we really are allocating across absolutely everything, and it’s just everyone in the industry right now.” “You used to be able to just scale up on demand into CPU capacity as long as your budget allowed you to. Right? And now it becomes something that you have to capacity plan for in advance, which is, you know, kind of punching everyone in the face. It’s brand new and people are starting to do it.” “And so dealing with that means a few things. One is obviously just go try to get more CPU. Cool. We’re doing that as much as we can. But two is we wanna be able to, at the infrastructure layer, run it more efficiently. Right?”
Sequoia’s Konstantine Buhler reveals the firm incubated an agentic AI CPU company with the team behind Apple Silicon and Qualcomm Snapdragon, calling CPUs a constraint that endures consistently while memory may be in a super cycle “So we have this insight, this series of insights. We know there’s constraints about compute. We realize that the particularly intense constraint is gonna be probably memory and then probably CPUs." "We say memory might be involved in a super cycle, maybe not something we’re gonna invest in right now. CPUs is something that endures consistently. Let’s go find the greatest team in the world to do this." "And that’s exactly what we did. So we worked hard to find the team that was critical in Apple Silicon and then critical in Qualcomm Snapdragon. And then they decided to leave Qualcomm. We approached them right after they left and said, would you think about building an agentic AI CPU, they said, yes." "We incubated the company. We put it together. We got it off to a great start about eight months ago. And we are on our way to having the AI agentic CPU, which now the entire industry has acknowledged is a major bottleneck, because we’ve gone from having one CPU for maybe 30 GPUs to a much higher ratio of that." "So that’s an example of how we have these conversations. We’ve got this debate. We say, what are we going to do? We’re going to turn this into monetization. And that team is doing amazing work.”
13
21
7
158
23,779
Congratulations @dMatrix_AI for your "Start-Up to Watch" nomination by @GlobalSemi for the 2026 Annual #GSA Awards
🚨 𝐓𝐇𝐄 𝐌𝐎𝐌𝐄𝐍𝐓 𝐖𝐄’𝐕𝐄 𝐁𝐄𝐄𝐍 𝐖𝐀𝐈𝐓𝐈𝐍𝐆 𝐅𝐎𝐑! The 2026 GSA Annual Awards nominees are officially revealed! For 30+ years, these awards have recognized the visionaries, innovators, and industry leaders shaping semiconductors. From bold startups to global industry giants, meet this year's nominees: gsaglobal.org/global-semicon… Congratulations to all! We look forward to celebrating you December 10th in Santa Clara. #GSA #semiconductor
1
8
338
Michael Stewart retweeted
Today we’re releasing the GLiDE family: five open-weight decision models, from 0.4B to 27B. Together, they set a new Pareto frontier for decision models. Most decisions inside an app don’t need a model that thinks out loud. They need a fast, clear, and calibrated answer. That's why we built GLiDE. The part I’m most excited about is the recurrent depth in 27B and 2B models. Selected middle layers run twice, giving the model more computation before it answers without adding parameters. On the Hugging Face Decision Index, GLiDE 27B is expected to score 73.3 vs. KnowLine 31B's 71.86, a 2% higher score with approximately ~11% fewer served params. All five models use the System One schema. Test the same tasks across sizes, evaluate them on your own workload, and run the smallest model that meets your needs.
22
28
4
257
201,143